Scalable FBP Decomposition for Cone-Beam CT Reconstruction

We use ABCI (https://abci.ai/) HPC to solve image reconstruction by FBP algorithm. This repository contains the artifact, e.g. job scripts, benchmarks.

Dependencies

We tested on ABCI, using Nvidia Volta V100 GPUs, with GCC 4.8.5 and NVCC 10.2 The following libraries and tools are requirements:

cmake = 3.1
CUDA >= 10.0
python >= 2.7
Malab R2018a
Intel MPI 2020.4.304
Intel IPP 2020.4.304
Insight Segmentation and Registration Toolkit (ITK)
Reconstruction Toolkit (RTK)

Compiler

g++ >= 4.8.5
nvcc >= 10.0
intel mpicc 2020.4.304

How to build

git clone git@github.com:peng-ac/dfbp-artifacts-sc21.git
cd dfbp; make clean all;

Note: all modules will be generated in dfbp/Release folder

Dataset

All geometric parameters used in our evaluation are presented in file geometry.sh.

Tomobank

https://tomobank.readthedocs.io/en/latest/#
we use tomo_00027, tomo_00028, tomo_00029, tomo_00030 in our evaluation

Coffee bean
```
private dataset
```
Bumblebee
```
private dataset
```

shepp-logan phantom

  script                 : phantom3d/phantom3d.m
  projections generation : ./generate-projections.sh

Jobs running on ABCI

all jobs used are stored in folder abci.jobs

Strong scaling evaluation jobs

set.Bumblebee.0008.sh
set.Bumblebee.0016.sh
set.Bumblebee.0032.sh
set.Bumblebee.0064.sh
set.Bumblebee.0128.sh
set.Bumblebee.0256.sh
set.Bumblebee.0512.sh
set.Bumblebee.1024.sh
set.coffeebean_1x.0016.sh
set.coffeebean_1x.0032.sh
set.coffeebean_1x.0064.sh
set.coffeebean_1x.0128.sh
set.coffeebean_1x.0256.sh
set.coffeebean_1x.0512.sh
set.coffeebean_1x.1024.sh
set.coffeebean_1x.2048.sh
set.coffeebean_2x.0004.sh
set.coffeebean_2x.0008.sh
set.coffeebean_2x.0016.sh
set.coffeebean_2x.0032.sh
set.coffeebean_2x.0064.sh
set.coffeebean_2x.0128.sh
set.coffeebean_2x.0256.sh
set.coffeebean_2x.0512.sh
set.coffeebean_2x.1024.sh
set.tomobank_29.0004.sh
set.tomobank_29.0008.sh
set.tomobank_29.0016.sh
set.tomobank_29.0032.sh
set.tomobank_29.0064.sh
set.tomobank_29.0128.sh
set.tomobank_29.0256.sh
set.tomobank_29.0512.sh
set.tomobank_29.1024.sh

Weak scaling evaluation jobs

set.Bumblebee.0128.sh
set.Bumblebee.0256.sh
set.Bumblebee.0512.sh
set.Bumblebee.1024.sh
set.coffeebean_1x.0064.sh
set.coffeebean_1x.0128.sh
set.coffeebean_1x.0256.sh
set.coffeebean_1x.0512.sh
set.coffeebean_1x.1024.sh
set.coffeebean_2x.0128.sh
set.coffeebean_2x.0256.sh
set.coffeebean_2x.0512.sh
set.coffeebean_2x.1024.sh
set.tomobank_29.0256.sh
set.tomobank_29.0512.sh
set.tomobank_29.1024.sh

How to run

the command we use on ABCI will be like qsub -g GROUP_ID job_name

example:  qsub -g GROUP_ID set.Bumblebee.0008.sh

Run benchmarks

The related benchmarks can be found in follows:

https://www.openrtk.org/

https://github.com/LLNL/ior

https://docs.nvidia.com/cuda/cuda-samples/index.html

Help/Support:

For more information or questions, contact the authors at peng.chen.ac#gmail.com (replace # by @, please)